DeepSCEM/patch.py at main · Cyril-Meyer/DeepSCEM · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
# this code is based on the project "Neural Networks Image Segmentation toolkit" : https://github.com/Cyril-Meyer/NeNISt
from random import randint, random
import numpy as np


def check_valid(image, label):
    if type(image) is list and type(label) is list:
        if not len(image) == len(label):
            return False
        for i in range(len(image)):
            if not image[i].shape[:-1] == label[i].shape[:-1] and len(image[i].shape) == 4:
                return False
    elif type(image) is np.ndarray and type(label) is np.ndarray:
        return image.shape[:-1] == label.shape[:-1] and len(image.shape) == 4
    else:
        return False
    return True


def create_label_indexes(label, patch_size):
    label_indexes = []
    if len(patch_size) == 2:
        pz = 0
        py, px = patch_size
    elif len(patch_size) == 3:
        pz, py, px = patch_size
    else:
        raise ValueError

    for i in range(label.shape[-1]):
        mz, my, mx, _ = label.shape
        label_indexes.append(np.argwhere(label[:mz - pz, :my - py, :mx - px, i] > 0))
    return label_indexes


def gen_patch_2d_batch(patch_size, image, label, batch_size,
                       augmentation_rotation, augmentation_flip,
                       label_indexes, label_indexes_prop):
    n_channel = image[0].shape[-1]
    n_label = label[0].shape[-1]
    image_dtype = image[0].dtype
    label_dtype = label[0].dtype
    patch_size_y, patch_size_x = patch_size

    batch_image = np.zeros((batch_size,) + tuple(patch_size) + (n_channel,), dtype=image_dtype)
    batch_label = np.zeros((batch_size,) + tuple(patch_size) + (n_label,), dtype=label_dtype)
    img = image
    lbl = label
    if label_indexes is not None:
        lbli = label_indexes
    while True:
        batch_image.fill(0)
        for i in range(batch_size):
            if type(image) is list:
                n = randint(0, len(image) - 1)
                img = image[n]
                lbl = label[n]
                if label_indexes is not None:
                    lbli = label_indexes[n]

            x = randint(0, img.shape[2] - patch_size_x)
            y = randint(0, img.shape[1] - patch_size_y)
            z = randint(0, img.shape[0] - 1)

            if label_indexes is not None and label_indexes_prop > random():
                cla = randint(0, len(lbli) - 1)
                if len(lbli[cla]) > 0:
                    r = randint(0, len(lbli[cla]) - 1)
                    z, y, x = lbli[cla][r]
                    y = np.clip(y - (patch_size_y // 2), 0, img.shape[1] - patch_size_y)
                    x = np.clip(x - (patch_size_x // 2), 0, img.shape[2] - patch_size_x)
                    z, y, x = int(z), int(y), int(x)

            batch_image[i, :, :, :] = img[z, y:y + patch_size_y, x:x + patch_size_x, :]
            batch_label[i, :, :, :] = lbl[z, y:y + patch_size_y, x:x + patch_size_x, :]

            if augmentation_rotation:
                if patch_size_y == patch_size_x:
                    rot = randint(0, 3)
                    batch_image[i, :, :] = np.rot90(batch_image[i, :, :], rot)
                    batch_label[i, :, :] = np.rot90(batch_label[i, :, :], rot)

            if augmentation_flip:
                if randint(0, 1) == 1:
                    batch_image[i, :, :] = np.fliplr(batch_image[i, :, :])
                    batch_label[i, :, :] = np.fliplr(batch_label[i, :, :])

                if randint(0, 1) == 1:
                    batch_image[i, :, :] = np.flipud(batch_image[i, :, :])
                    batch_label[i, :, :] = np.flipud(batch_label[i, :, :])

        yield batch_image, batch_label


def gen_patch_3d_batch(patch_size, image, label, batch_size,
                       augmentation_rotation, augmentation_flip,
                       label_indexes, label_indexes_prop):
    n_channel = image[0].shape[-1]
    n_label = label[0].shape[-1]
    image_dtype = image[0].dtype
    label_dtype = label[0].dtype
    patch_size_z, patch_size_y, patch_size_x = patch_size

    batch_image = np.zeros((batch_size,) + tuple(patch_size) + (n_channel,), dtype=image_dtype)
    batch_label = np.zeros((batch_size,) + tuple(patch_size) + (n_label,), dtype=label_dtype)
    img = image
    lbl = label
    if label_indexes is not None:
        lbli = label_indexes
    while True:
        batch_image.fill(0)
        for i in range(batch_size):
            if type(image) is list:
                n = randint(0, len(image) - 1)
                img = image[n]
                lbl = label[n]
                if label_indexes is not None:
                    lbli = label_indexes[n]

            x = randint(0, img.shape[2] - patch_size_x)
            y = randint(0, img.shape[1] - patch_size_y)
            z = randint(0, img.shape[0] - patch_size_z)

            if label_indexes is not None and label_indexes_prop > random():
                cla = randint(0, len(lbli) - 1)
                if len(lbli[cla]) > 0:
                    r = randint(0, len(lbli[cla]) - 1)
                    z, y, x = lbli[cla][r]
                    z = np.clip(z - (patch_size_z // 2), 0, img.shape[0] - patch_size_z)
                    y = np.clip(y - (patch_size_y // 2), 0, img.shape[1] - patch_size_y)
                    x = np.clip(x - (patch_size_x // 2), 0, img.shape[2] - patch_size_x)
                    z, y, x = int(z), int(y), int(x)

            batch_image[i, :, :, :, :] = img[z:z + patch_size_z, y:y + patch_size_y, x:x + patch_size_x, :]
            batch_label[i, :, :, :, :] = lbl[z:z + patch_size_z, y:y + patch_size_y, x:x + patch_size_x, :]

            if augmentation_rotation:
                if patch_size_z == patch_size_x:
                    rot = randint(0, 3)
                    batch_image[i, :, :] = np.rot90(batch_image[i, :, :, :], rot, axes=(0, 2))
                    batch_label[i, :, :] = np.rot90(batch_label[i, :, :, :], rot, axes=(0, 2))
                if patch_size_z == patch_size_y:
                    rot = randint(0, 3)
                    batch_image[i, :, :] = np.rot90(batch_image[i, :, :, :], rot, axes=(0, 1))
                    batch_label[i, :, :] = np.rot90(batch_label[i, :, :, :], rot, axes=(0, 1))
                if patch_size_y == patch_size_x:
                    rot = randint(0, 3)
                    batch_image[i, :, :] = np.rot90(batch_image[i, :, :, :], rot, axes=(1, 2))
                    batch_label[i, :, :] = np.rot90(batch_label[i, :, :, :], rot, axes=(1, 2))

            if augmentation_flip:
                if randint(0, 1) == 1:
                    batch_image[i, :, :, :] = np.flip(batch_image[i, :, :, :], 0)
                    batch_label[i, :, :, :] = np.flip(batch_label[i, :, :, :], 0)

                if randint(0, 1) == 1:
                    batch_image[i, :, :, :] = np.flip(batch_image[i, :, :, :], 1)
                    batch_label[i, :, :, :] = np.flip(batch_label[i, :, :, :], 1)

                if randint(0, 1) == 1:
                    batch_image[i, :, :, :] = np.flip(batch_image[i, :, :, :], 2)
                    batch_label[i, :, :, :] = np.flip(batch_label[i, :, :, :], 2)

        yield batch_image, batch_label


def gen_patch_batch(patch_size, image, label, batch_size=32, augmentation_rotation=False, augmentation_flip=True, label_indexes=None,
                    label_indexes_prop=1.0):
    gen = None
    if not (len(patch_size) == 2 or len(patch_size) == 3):
        raise ValueError
    if not check_valid(image, label):
        raise ValueError
    if len(patch_size) == 2:
        gen = gen_patch_2d_batch(patch_size, image, label, batch_size,
                                 augmentation_rotation, augmentation_flip,
                                 label_indexes, label_indexes_prop)
    elif len(patch_size) == 3:
        gen = gen_patch_3d_batch(patch_size, image, label, batch_size,
                                 augmentation_rotation, augmentation_flip,
                                 label_indexes, label_indexes_prop)
    else:
        raise ValueError
    return gen


def crop_gen_patch_2d_batch(gen, crop_in, crop_out):
    while True:
        X, Y = next(gen)
        y_min = crop_in[0][0]
        y_max = X.shape[1] - crop_in[0][1]
        x_min = crop_in[1][0]
        x_max = X.shape[2] - crop_in[1][1]
        X = X[:, y_min:y_max, x_min:x_max]

        y_min = crop_out[0][0]
        y_max = Y.shape[1] - crop_out[0][1]
        x_min = crop_out[1][0]
        x_max = Y.shape[2] - crop_out[1][1]
        Y = Y[:, y_min:y_max, x_min:x_max]

        yield X, Y


def gen_to_multiple_outputs(gen):
    while True:
        X, Y = next(gen)

        Y_ = []
        for c in range(Y.shape[-1]):
            Y_.append(Y[..., c:c + 1])

        yield X, Y_